Curated notes on external repos that complement Cursor, Claude Code, and vibe coding workflows.
Most entries are link-only; two Vercel SKILL.md summaries are also vendored under cursor/skills/ (see the Vercel section below). All descriptions reflect public READMEs and repository layout as of the exploration date.
| Repo | Upstream |
|---|---|
| Vercel Agent Skills | vercel-labs/agent-skills |
| Code Review Graph | tirth8205/code-review-graph |
| Awesome DESIGN.md | VoltAgent/awesome-design-md |
| claude-md (hooks + CLAUDE.md) | iamfakeguru/claude-md |
| Claw Code | ultraworkers/claw-code |
| Agentic AI prompt research | Leonxlnx/agentic-ai-prompt-research |
| HumanLayer / CodeLayer | humanlayer/humanlayer |
| Harness (agent teams) | revfactory/harness |
License: MIT (per upstream README).
What it is: Official Vercel collection of Agent Skills (packaged instructions; optional scripts). Skills follow the Agent Skills format (SKILL.md, optional scripts/, references/).
Layout explored: skills/ contains topical packages, including:
react-best-practices— React / Next.js performance rulesweb-design-guidelines— UI review (accessibility, UX, performance)react-native-guidelines— React Native / Exporeact-view-transitions— View Transition API and Next.jscomposition-patterns— scalable React compositiondeploy-to-vercel— claimable deployments from agent chatsvercel-cli-with-tokens— CLI usage with tokens
Use here: Install upstream (optional; keeps you aligned with latest skill packaging):
npx skills add vercel-labs/agent-skillsVendored in this hub (summary SKILL.md only): web-design-guidelines, react-best-practices. Detailed React rule files (rules/*.md) and compiled AGENTS.md are not copied; fetch from the upstream paths linked in those skills.
Fit for this hub: Complements cursor/skills/README.md and cursor/best-practices/ for frontend and UI audits.
What it is: Local knowledge graph for Claude Code: persistent map of the codebase so reviews and daily work ingest fewer irrelevant tokens (upstream claims large token reductions; verify in your own projects).
Layout explored (root): .mcp.json, CLAUDE.md, CHANGELOG.md, .github/ — typical Claude Code + MCP integration layout.
Use here: Clone or follow upstream install in a project where you want graph-backed context. Keep this repo as a pointer only unless you decide to submodule or document a specific setup.
Fit for this hub: Pairs with cursor/commands/code-review.md and code-review workflows; advanced context engineering.
License: MIT (per upstream README).
What it is: Curated DESIGN.md files (Stitch-style design system descriptions) derived from public sites, plus preview.html / preview-dark.html per theme under design-md/<site>/.
Use here: Copy a single site's DESIGN.md into a product prototype and ask your agent to match UI (upstream workflow). Do not mirror the whole collection in this repo.
Fit for this hub: Strong match for Product Builders and designers — see main README audience.
What it is: Production-style CLAUDE.md plus shell hooks under .claude/hooks/ (verify after edits, block destructive commands, truncation checks, etc.). Addresses common Claude Code failure modes via mechanical hooks.
Layout explored: CLAUDE.md, install.sh, .claude/ (hooks, settings).
Use here: Prefer upstream install in your own project:
curl -sL https://raw.githubusercontent.com/iamfakeguru/claude-md/main/install.sh | bash -s .Review hook behavior before running; adapt commands to your stack (TypeScript vs Python, etc.).
Fit for this hub: Complements cursor/security/ and team discipline docs; does not replace reading Anthropic’s own Claude Code docs.
What it is: Public Rust implementation of the claw CLI agent harness; canonical code in rust/, with USAGE.md, PARITY.md, PHILOSOPHY.md. Companion Python/reference under src/ and tests/.
Use here: Build from rust/ if you want a standalone CLI; this hub stays documentation-only unless you add an explicit submodule later.
Fit for this hub: Ecosystem reference next to openclaw/README.md (different project; name similarity only).
What it is: Educational reconstructed prompt patterns for agentic coding assistants (orchestration, sub-agents, permissions, compaction, skills). Files live in prompts/.
Disclaimer (upstream): Reconstructions from observation — not official leaks.
Use here: Cross-read with prompt-engineering/README.md. Link only; do not treat as vendor truth.
Fit for this hub: Research track alongside cursor/references/agentic-patterns.md.
What it is: Monorepo for CodeLayer (agent orchestration IDE built on Claude Code) and the HumanLayer SDK; Turbo/Bun workspace, Go and TypeScript, docs under docs/.
Use here: Follow upstream releases and README.md for install / waitlist. Not vendored here.
Fit for this hub: “Scale agents across a team” angle — useful for advanced readers exploring beyond Cursor-only flows.
License: Apache-2.0 (per upstream README).
What it is: Claude Code plugin / meta-skill that designs agent teams, generates .claude/agents/ and .claude/skills/ for a domain. Includes skills/harness/SKILL.md and reference markdown under references/.
Install (upstream): Marketplace flow documented in their README (/plugin marketplace add, /plugin install).
Use here: Trigger prompts like “Build a harness for this project” inside Claude Code after install.
Fit for this hub: Extends multi-agent thinking beyond cursor/references/agentic-patterns.md.
When adding new upstream repos to this hub: extend this catalog, then link it from references/README.md and from nice-projects/README.md if the entry is also a “project to watch.”